An Automatic Algorithm Configuration based on a Bayesian Network

Autor(es)

do Nascimento, Marcelo Branco ; Chaves, Antonio Augusto ; IEEE

• Tipo de documento: Artigo Científico
• Fonte: 2020 IEEE CONGRESS ON EVOLUTIONARY COMPUTATION (CEC); v. N/A, p. 8-pg., 2020-01-01.

Resumo

The parameter tuning process is one of the main tasks in the development of metaheuristics. The appropriate parameter can assist in finding good solutions to combinatorial optimization problems. However, finding a good parameter setting is a hard task. It involves understanding how the relationship between parameters connects to the problem scenario. This article proposes a method to automatically tune parameters of metaheuristics, called the Bayesian Network Tuning (BNT). Our goal is to develop an efficient method in terms of solution quality and computational time, which can find configurations that support metaheuristics in solving optimization problems. In order to evaluate this method, a Biased Random-Key Genetic Algorithm (BRKGA) was implemented to solve the Bin Packing Problem. The BRKGA was tuned with our method and other tuning methods found in the literature. A comparison of the results shows that the proposed method found good solutions and was competitive in relation to the other tuning methods. (AU)

• Processo FAPESP: 18/15417-8 - Desenvolvimento de uma meta-heurística híbrida com fluxo de controle e parâmetros adaptativos
• Beneficiário: Antônio Augusto Chaves
• Modalidade de apoio: Auxílio à Pesquisa - Jovens Pesquisadores - Fase 2